ML Engineer II - Conversation Memory & Retrieval

Warmada; United States; Remote; USD 129,000 - 170,000; Full time

Benefits offered by this job

Health care insurance

401(k) retirement account

Paid time off

Job summary

Twilio is hiring a Machine Learning Engineer, Customer Memory, to build scalable pipelines that convert raw conversation data into structured memory for fast retrieval by agents. You will join the Conversation Memory team, focusing on quality and accuracy as the product evolves.

You will own the extraction and retrieval components, work with embeddings and vector search, and contribute to evaluation systems to ensure memory accuracy across millions of conversations.

Qualifications

• Experience building backend services and picking up new languages and tools.
• Curious about production systems, operating them with rigor and accountability.
• Experience with evaluation/quality-measurement systems or willingness to learn.
• Strong observability practices: logging, metrics, tracing.
• Fluency with AI-assisted dev tools applied with critical thinking.

Responsibilities

• Build the extraction pipeline turning raw data into structured memory.
• Own the retrieval path using embeddings, vector search, and ranking.
• Contribute to eval systems with benchmarks and regression gates.
• Diagnose production issues across ingestion, extraction, and indexing paths.
• Partner with Product and ML teams to translate problems into system changes.

Skills

Backend services

Production observability

Go

Python

AI-assisted dev tools

LLM-backed products

Evaluation systems

Kafka

Embeddings/vector search

Tools

Kafka

Embeddings

Vector search

Job description

Twilio is hiring a Machine Learning Engineer, Customer Memory, to build scalable pipelines that convert raw conversation data into structured memory for fast retrieval by agents. You will join the Conversation Memory team, focusing on quality and accuracy as the product evolves.

You will own the extraction and retrieval components, work with embeddings and vector search, and contribute to evaluation systems to ensure memory accuracy across millions of conversations.